The present invention relates to an expression library comprising a repertoire of nucleic acid sequences derived from the natural sequence repertoire but modified to enhance the extent of sequence variability, each said nucleic acid sequence encoding at least part of a variable domain of a heavy chain derived from an immunoglobulin naturally devoid of light chains and its use in producing antibodies, or more particularly fragments thereof.
In particular, the invention relates to a method for the preparation of antibodies or fragments thereof, having binding specificity for a target antigen which avoids the need for the donor previously to have been immunised with the target antigen.
Monoclonal antibodies, or binding fragments thereof, have traditionally been prepared using hybridoma technology (Kohler and Milstein, 1975, Nature 256, 495). More recently, the application of recombinant DNA methods to generating and expressing antibodies has found favour. In particular, interest has concentrated on combinatorial library techniques with the aim of utilizing more efficiently the antibody repertoire.
The natural immune response in vivo generates antigen-specific antibodies via an antigen-driven recombination and selection process wherein the initial gene recombination mechanism generates low specificity, low-affinity antibodies. These clones can be mutated further by antigen-driven hypermutation of the variable region genes to provide high specificity, high affinity antibodies.
Approaches to mimicking the first stage randomisation process which have been described in the literature include those based on the construction of xe2x80x98naivexe2x80x99 combinatorial antibody libraries prepared by isolating panels of immunoglobulin heavy chain variable (VH) domains and recombining these with panels of light variable chains (VL) domains (see, for example, Gram et al, Proc. Natl. Acad. Sa, USA, 89, 3576-3580, 1992). Naive libraries of antibody fragments have been constructed, for example, by cloning the rearranged V-genes from the IgM RNA of B cells of un-immunised donors isolated from peripheral blood lymphocytes, bone marrow or spleen cells (see, for example, Griffiths et al, EMBO Journal, 12(2), 725-734, 1993, Marks et al, J. Mol. Biol., 222, 581-597, 1991). Such libraries can be screened for antibodies against a range of different antigens.
In combinatorial libraries derived from a large number of VH genes and VL genes, the number of possible combinations is such that the likelihood that some of these newly formed combinations will exhibit antigen-specific binding activity is reasonably high provided that the final library size is sufficiently large.
Given that the original B-cell pairings between antibody heavy and light chain, selected by the immune system according to their affinity of binding, are likely to be lost in the randomly, recombined repertoires, low affinity pairings would generally be expected. In line with expectations, low affinity antibody fragments (Fabs) with Kas of 104-105 Mxe2x88x921 for a progesterone-bovine serum albumin (BSA) conjugate have been isolated from a small (5xc3x97106) library constructed from the bone marrow of non-immunised adult mice (Gram et al, see above).
Antibody fragments of higher affinity (Kas of 106-107 Mxe2x88x921 range) were selected from a repertoire of 3xc3x97107 clones, made from the peripheral blood lymphocytes of two healthy human volunteers (Marks et al, see above) comprising heavy chain repertoires of the IgM (naive) class. These were combined with both Lamda and Kappa light chain sequences, isolated from the same source. Antibodies to more than 25 antigens were isolated from this library, including self-antigens (Griffiths et al, see above) and cell-surface molecules (Marks et al, Bio/Technology, 11, 1145-1149, 1993).
The second stage of the natural immune response, involving affinity maturation of the selected specificities by mutation and selection has been mimicked in-vitro using the technique of random point mutation in the V-genes and selecting mutants for improved affinity.
Recently, the construction of a repertoire of 1.4xc3x971010 scfv clones, achieved by xe2x80x98brute forcexe2x80x99 cloning of rearranged V genes of all classes from 43 non-immunised human donors has been reported (Vaughan et al 1996) and Griffiths et al, see above. Antibodies to seven different targets (including toxic and immunosuppressant molecules) were isolated, with measured affinities all below 10 nM.
The main limitation in the construction of combinatorial libraries is their size, which consequently limits their complexity. Evidence from the literature suggests that there is a direct link between library size and diversity and antibody specificity and affinity (see Vaughan., et al, Nature Biotechnology, 14, 309-314, 1996), such that the larger (and more diverse) the library, the higher the affinity of the selected antibodies.
The optimisation of binding affinity through random recombination of a heavy and light chain in combinatorial libraries is complicated by sequence variations in the two framework regions, (i.e. the parts of the variable domains that serve as a scaffold in supporting the regions of hypervariability which are in turn termed the complementary determining regions or CDRs.
Only some combinations of framework sequences are compatible with the folding and interaction required for the correct orientation of the 6 CDRs that is necessary for good binding affinity. Consequently, conventional combinatorial libraries are likely to contain a high percentage of molecules that are non-functional.
The affinity of antibodies may also be improved by the process of xe2x80x9cchain shufflingxe2x80x9d, whereby a single heavy or light chain is recombined with a library of partner chains (Marks et al, Bio/Technology, 10, 779-782, 1992).
EP-A-0368684 (Medical Research Council) discloses the construction of expression libraries comprising a repertoire of nucleic acid sequences each encoding at least part of an immunoglobulin variable domain and the screening of the encoded domains for binding activities. It is stated that repertoires of genes encoding immunoglobulin variable domains are preferably prepared from lymphocytes of animals immunised with an antigen.
The isolation of single VH domains having antigen binding activities, facilitated by immunisation, is exemplified (see Example 6).
These results illustrate that although the VH part alone of a classical antibody binding domain can exhibit binding activity, the specificity and affinity are generally very low. This may be explained by the absence of the functional involvement of the missing light chain such that only half of the intended binding pocket is present, leading to binding with related or homologous targets.
EP-A-0368684 further describes the cloning of heavy chain variable domains with binding activities generated by mutagenesis of one or each of the CDRs. The preparation of a repertoire of CDR3s is described by using xe2x80x9cuniversalxe2x80x9d primers based in the flanking sequences, and likewise repertoires of other CDRs singly or in combination. These synthetic mutant VH clones can then be recombined with VL chains to produce a synthetic combinatorial library.
Construction of libraries by such synthetic recombinatorial techniques produces a repertoire of molecules that collectively exhibit a good degree of binding diversity, wherein the variability is focused into the sequences that encode the CDRs of each chain. However this technique does not overcome the problems previously discussed with respect to random recombination of heavy and light chains and production of non-functional molecules. Furthermore there are still six separate regions (3 CDRs in VH and another 3 in VL) determining the binding capacity of the molecule hence the repertoire of possible binding variants is encoded within a rather diffuse stretch of coding sequence, thus making a focused approach to altering the binding affinity of these binding domains very difficult.
There remains a continuing need for the development of improved methods for constructing libraries of immunoglobulin binding domains. In particular, it would be desirable to avoid the recombination of heavy and light chains thereby preventing the formation of molecules that are non-functional following recombination.
It would also be an advantage to reduce the number of hypervariable residues in the binding domain as this would allow a more complete repertoire of possible binding variants to be obtained.
WO 94/4678, Casterman et al, describes immunoglobulins capable of exhibiting the functional properties of conventional (four-chain) immunoglobulins but which comprise two heavy polypeptide chains and which furthermore are devoid of light polypeptide chains. Fragments of such immunoglobulins, including fragments corresponding to isolated heavy chain variable domains or to heavy chain variable domain dimers linked by the hinge disulphide are also described. Methods for the preparation of such antibodies or fragments thereof on a large scale comprising transforming a mould or yeast with an expressible DNA sequence encoding the antibody or fragment are described in patent application WO 94/25591 (Unilever).
The immunoglobulins described in WO 94/4678, which may be isolated from the serum of Camelids, do not rely upon the association of heavy and light chain variable domains for the formation of the antigen-binding site but instead the heavy polypeptide chains alone naturally form the complete antigen binding site. These immunoglobulins, hereinafter referred to as xe2x80x9cheavy-chain immunoglobulinsxe2x80x9d are thus quite distinct from the heavy chains derived from conventional (four-chain) immunoglobulins. Heavy chains from conventional immunoglobulins contribute part only of the antigen-binding site and require a light chain partner, forming a complete antigen binding site, for optimal antigen binding.
As described in WO 94/4678, heavy chain immunoglobulin VH regions isolated from Camelids (hereinafter VHH domains) which form a complete antigen binding site and thus constitute a single domain binding site differ from the VH regions derived from conventional four-chain immunoglobulins in a number of respects, notably in that they have no requirement for special features for facilitating interaction with corresponding light chain domains. Thus, whereas in conventional (four-chain) immunoglobulins the amino acid residue involved in the VH/VL interaction is highly conserved and generally apolar leucine, in Camelid derived VH domains this is replaced by a charged amino acid, generally arginine. It is thought that the presence of charged amino acids at this position contributes to increasing the solubility of the camelid derived VH. A further difference which has been noted is that one of the CDRs of the heavy chain immunoglobulins of WO 94/4678, the CDR3, may contain an additional cysteine residue associated with a further additional cysteine residue elsewhere in the variable domain. It has been suggested that the establishment of a disulphide bond between the CDR3 and the remaining regions of the variable domain could be important in binding antigens and may compensate for the absence of light chains.
cDNA libraries composed of nucleotide sequences coding for a heavy-chain immunoglobulin and methods for their preparation are disclosed in WO 94/4678. It is stated that these immunoglobulins have undergone extensive maturation in vivo and the V region has naturally evolved to function in the absence of the light chain variable domain. It is further suggested that in order to allow for the selection of antibodies having specificity for a target antigen, the animal from which the cells used to prepare the library are obtained should be pre-immunised against the target antigen. No examples of the preparation of antibodies are given in the specification of WO 94/4678. The need for prior immunisation is also referred to in Arabi Ghahroudi et al (FEBS Letters, 414, (1997), 521-526).
Davies et al (Bio/Technology, 13, 475-479,. 1995) describe an approach to the construction of a library of binding domains based on a modified human VH domain which is intended to mimic a camelid VHH domain. This method involves replacement of the sequence segment encoding one of the CDRs by random, synthetic sequences. Although it was possible to isolate domains with selected antigen binding properties from the resulting library, these were generally characterised by poor binding affinity and specificity for protein antigens. The results would not seem, therefore, to recommend the further application of this type of approach.
The present invention relates to an expression library comprising a repertoire of synthetic or semi-synthetic nucleic acid sequences, not cloned from an immunised source, wherein said nucleic acid sequences are derived from immunoglobulins that are naturally devoid of light chains.
This invention is based on the unexpected finding that high affinity, high specificity antibodies or fragments thereof capable of binding either to protein or small molecule antigens can be obtained from a non-immunised camelid source provided that random mutagenesis of one or more CDRs is carried out, or that alternative combinations of existing CDRs are generated, in order to increase the extent of sequence variability in the antibody repertoire.
The present invention therefore provides an expression library comprising a repertoire of nucleic acid sequences, which sequences are not cloned from an immunised source, each nucleic acid sequence encoding at least part of a variable domain of a heavy chain derived from an immunoglobulin naturally devoid of light chains wherein the extent of sequence variability in said library is enhanced compared to the corresponding naive expression library by introducing mutations in one or more of the complementarity determining regions (CDRS) of said nucleic acid sequences or by random recombination of fragments of said nucleic acid sequences, thereby generating alternative combinations of CDR and framework sequences not naturally present in the naive library repertoire.